OSC Fall Update • Budget Trends • SOCC Charges • Business Model Proposal • Discussion
Budget Trends – Revenue Analysis OSC Revenue Categories: Revenue Trend – State Subsidy – Sponsored Research Funding – User Fees • Commercial Usage • Condo (Dedicated) Usage 2018 Breakdown 11% State Line Item 10% Appropriations Sponsored Research Funding 2013 2014 2015 2016 2017 2018 User Fees (Budget) 79%
Budget Trends – Expense Analysis OSC Expense Categories: Expense Trend – Direct Costs – Indirect Costs – Personnel 2018 Breakdown Personnel Space Rent and Power 11% 2% 3% Equipment Maint. and 0% Purchased Services Equipment 5% Other Direct Expenditures Overhead / Indirect 2013 2014 2015 2016 2017 2018 F&A (Budget) Cost Distributions and 54% 25% Transfers
Budget Trends – Overall OSC Budget Trend Projected funding gap is a result of: Decreasing Funding Increasing Expenses 2013 2014 2015 2016 2017 2018 (Budget) Revenue Expense
Impact of SOCC Charges State Subsidy Trend $5 Core Operations Space and Power $4 10% Decrease $3 Millions in FY18 $2 $1 $0 – Data center costs fully transferred to OSC by 2017 – Permanent, substantial addition to OSC’s expenses – From 2008 to 2018, OSC lost nearly $1.4M for core operations
OSC Business Model Proposal • OSC is an initiative of the Ohio Department of Higher Education (ODHE) • Decision was made to ask universities for support – Need a single policy to apply to all Ohio universities • ODHE has experience with fee models for services (e.g. OARnet, OhioLink), wants to do the same thing with OSC – Faculty chargeback policies up to each university
OSC Business Model Proposal Features • Maintain subsidized access model – Subject to peer review by OSC Allocations Committee – Fully subsidize a portion of ALL awards • Charge fees based on compute (RU) usage – Options for institutions to pay centrally to lower per-RU fees – Establish predictable costs over multiple years • Create an advisory model similar to OARnet – User/Institutional-based Finance Committee – Each fall, Committee will develop pricing for coming fiscal year We are doing everything we can to constrain costs and will continue to offer heavily subsidized services that are well below market rates.
Business Model Proposal Annual Process Step 3: Subsidize 1 st 10K RU Step 5: Incentivize central payment Step 1: Project resource usage for all clients Available RUs (Millions) 40 Payment Discount Unit Cost 10% 90% $10,000 20% $0.136/RU 30 $30,000 40% $0.102/RU 20 $60,000 60% $0.068/RU 90% 10% 10 $350,000 80% $0.034/RU - Usage 2013 2014 2015 2016 2017 Step 6: Calculate cost/university Step 2: Quantify recovery target Step 4: Divide target by usage Institutional Costs ($Ks) RUs Unit Cost BGSU Expenses <10,000 No Cost UC >10,000 $0.17/RU OU Revenue Target CASE OSU DRAFT COSTS - 200 400 600 800
Alternatives to OSC for Example 100K RUs / Year Client Local Cluster Amazon Cloud Services • 1 Owens node ~= Amazon r4.8xlarge instance • 1 Owens node = ~22K RUs / year • 100K RUs = ~5 instances • 100K RUs = ~5 nodes • Typical instance upfront annual cost = ~$12K • Typical node upfront cost = ~$5,500 • 5 instance annual cost (paid upfront) = ~$60K • Typical node annual operating cost = ~$1,900 • 5 instance five-year lifetime cost = ~$300K • 5 node cluster cost = ~$28K upfront, • Additional costs for: ~$10K/year – data transfer • 5 node cluster five-year lifetime cost = $78K – system administration services • Additional costs for: – software licenses – data storage and backup services – system administration services – software licenses These costs are hard to quantify generically since – data storage and backup services each client has specific unique needs, although they are covered by OSC today.
Comparison to OSC for Example 100K RUs / Year Client Annual Central Annual Cost Institution RU Rate Payment for 100K RUs Does NOT include: Local Cluster $0 $0 $15,600 Amazon $0 $0 $60,000 • System administration • Software licenses Ohio State University $350,000 $0.034 $3,060 • Data storage Case Western Reserve University $60,000 $0.068 $6,120 • Backup services Ohio University $30,000 $0.102 $9,180 University of Cincinnati $30,000 $0.102 $9,180 Bowling Green State University $10,000 $0.136 $12,240 All Other Institutions $0 $0.170 $15,300 These are based upon projected institutional usage and lowest overall impact to each institution.
Client Feedback to Date Representative quotes: • "I would have to downgrade my research program to fit within the computation that I can afford, which is basically nothing at the rates that you are proposing.” • “Funding rates are plummeting, and the amount per grant is dropping in relative terms…If costs are pushed onto the grants that mainly support the salaries of young talent, the obvious outcome is that the workforce must be reduced.” Summary of comments: • Paying for compute from direct grant funds hurts ability to hire students • Concern over cost containment • Concern over long-term cost predictability • Desired flexibility to spend funding when available • Explore charging for other services (e.g. storage, GPUs, big-data, HIPPA)
Status and Next Steps • Discussions with provosts from 6 institutions this summer • DHE and OSC will continue to engage universities to discuss the proposed model and consider adjustments • OSC and Finance Committee will be looking at budget and usage projections • Begin charging in FY19 (July 1, 2018) with the goal of reaching $1.5M in funding by FY20
Questions?
Campus Visits • Bowling Green State University (January) • University of Cincinnati (April) • Ohio State University (March, June) – Consultation hours at Research Commons every other Tuesday
Upcoming Training and Events • Introduction to Supercomputing / Big Data at UC: October 10th • Big Data at OSC: October 26th See https://www.osc.edu/events
OSC Supercomputers + Storage
Owens Node Configurations “side-by-side” Comparison Node Type Compute GPGPU Data Analytics Node Count 648 160 16 Core Count 28 28 48 Core Type Broadwell Broadwell Haswell Memory 128 GB 128 GB 1500 GB Disk 1 TB 1 TB 20 TB GPU N/A P100 None
Owens GPU Rollout and Adoption • 160 Nvidia P100 GPUs were made available to clients in % Owens GPUs in Use the beginning of April 2017 100% 80% • Usage: 60% • ~50% Molecular dynamics (e.g. Amber, 40% Gromacs, Namd, 20% Lammps) 0% • ~50% Machine learning / April-17 May-17 June-17 July-17 neural networks (e.g. Tensorflow, Caffe, Torch)
System Monitoring
Client Portal Project • A user friendly client portal that is a “one ‐ stop” shop for registration, project management, profile updates, etc. • Status: – Internal user testing underway, external client testing beginning shortly – Development timeline expected to finish first week of October – AweSim website in process of being transferred to internal hosting
FY17-18 Roadmap of OSC Resources
New Cluster (C18) Vision / Timeline • Vision • Complement Owens • Dense compute component • GPU computer component • Big data component • Timeline • RFI issued Sept 25 • RFI due Oct 22 • RFP issued Nov 2 • RFP due Dec 15 • Facilities updates May • System delivery June
Recommend
More recommend
